01
Free
See Refario working with real traffic.
€0/month
- 10k runs / month
- 1 project
- 1 user
- 7-day retention
- Runs, traces, dashboards, and SDK ingestion
Operational visibility for AI systems
You built an AI system. Now see what it’s doing.
Refario connects every run, model, tool, euro, and failure—so the answer is never hidden across five dashboards.
No credit card First 10k runs free Setup in minutes
Fig. 01A complete request, not another summary
01Requestreceived18ms
02Agentrouted62ms
03Modelgenerated1.42s
04Toolretried2.84s
05Outputdelivered5.10s
Run cost€0.034+€0.009 retry
Latency5.10sTool path +2.41s
PolicyPassed6 / 6 rules
Works across the stack
OpenAIAnthropicGeminiCursorMCPSlackNotionCustom SDK
01 / The problem
AI systems rarely fail in one place. Your observability shouldn’t live in five.
A bad answer can begin with a model, a tool, a retry, a policy, or a routing decision. Refario keeps the whole path intact so you can understand the system—not just the last call.
01—05
The customer sees one answer. Refario sees the system that produced it.
A single slow response is enough to show the difference. Refario follows the request through its model, tools, retries, policy checks, cost, and final output.
See the product layerrun_6f93 / production5.10 seconds
01Customer asksWhere is my order?
02Agent routessupport-resolution
03Tool callsshipment.timeline.fetch
04Retry detected+2.41s · +€0.009
05Answer deliveredpolicy passed
Root causeTool retry
Cost impact+26%
Latency impact+2.41s
PolicyPassed
02 / One record
From first signal to clear cause.
Refario is built around the questions people actually ask when AI reaches production: what changed, where did it happen, what did it cost, and what should we fix first?
01
Runs
Every request, model call, tool, retry, and output on one readable timeline.
02
Cost
Spend attached to the workflow and customer behavior that created it.
03
Control
Guardrails and incidents visible beside the execution path they affect.
04
Tools
MCP and internal tool health without breaking the trace into another product.
03 / Shared evidence
One system, useful to everyone responsible for it.
Technical depth for builders. Cost context for decision makers. A shared operational record for everyone in between.
Engineering
- Find failed spans and retries
- Watch latency after releases
- Explain regressions with evidence
Product
- See real workflow outcomes
- Compare behavior by release
- Prioritize high-impact fixes
Finance
- Attribute spend by workflow
- Forecast AI cost
- Give budgets clear ownership
Operations
- Review guardrail coverage
- Monitor tool reliability
- Coordinate incident response
04 / Setup
Keep your stack. Add the missing operating layer.
Instrument a workflow, verify the connected path, and move from raw telemetry to a shared answer without rebuilding your application around Refario.
01
Connect
Send run, model, tool, and workflow events from the stack you already use.
02
Understand
Refario turns scattered telemetry into one operational record.
03
Improve
Fix failures, control cost, and report from evidence everyone can share.
05 / Pricing
Start with real traffic, not a sales call.
The free plan includes enough capacity to see how Refario works with your system. Upgrade when the operational value is already clear.
02
Starter
For one team operating a live AI product.
€19/month
- 100k runs / month
- 3 projects
- 2 users
- 30-day retention
- Anomaly detection, guardrails, and alerts
03
TeamRecommended
For shared AI operations across functions.
€59/month
- 1M runs / month
- 10 projects
- 10 users
- 90-day retention
- Finance dashboards and scheduled reports
04
Enterprise
For scale, compliance, and custom rollout needs.
Custom
- Higher negotiated limits
- SSO, audit logs, and SLA
- Dedicated onboarding
- Custom deployment options
06 / Compare
Choose the operating model, not the longest checklist.
Refario is the stronger fit when tracing is only the start of the question—and cost, tools, guardrails, and workflow ownership need to stay attached.
Langfuse
Langfuse alternative
For teams that need more than tracing once AI workflows hit production.
You need one operating view for workflow health, spend, guardrails, and MCP tools.
You want to move from a cost spike or incident to the exact workflow and runtime path that caused it.
LangSmith
LangSmith alternative
For teams that need production workflow operations beyond app development workflows.
You need a shared surface for engineering, AI platform, and budget review once traffic is live.
You care about cost per workflow, tool failures, and guardrail incidents as much as traces.
Helicone
Helicone alternative
For teams that need full workflow operations, not only request or gateway visibility.
Your runtime depends on multi-step workflows, MCP tools, and policy checks.
You need to connect provider costs and failures back to the workflow or incident they changed.
Arize Phoenix
Arize Phoenix alternative
For teams that want a tighter workflow-operations wedge in production.
You want a narrower product story around workflow operations and cost control.
Your team needs daily visibility into tool health, budget drift, and guardrail incidents.
Start with one project
Know what your AI system is doing.
No credit card. Connect real traffic and see the first operational record in minutes.